Class: SvmToolkit::Svm

Inherits:
Object
  • Object
show all
Defined in:
lib/svm_toolkit/svm.rb

Overview

Extends the Java SVM class

Available methods include:

Svm.svm_train(problem, param)

problem

instance of Problem

param

instance of Parameter

Returns an instance of Model

Svm.svm_cross_validation(problem, param, nr_folds, target)

problem

instance of Problem

param

instance of Parameter

nr_fold

number of folds

target

resulting predictions in an array

Defined Under Namespace

Classes: ContourDisplay, CrossValidationSearch, SvmTrainer

Class Method Summary collapse

Class Method Details

.cross_validation_search(training_set, cross_valn_set, costs = [-2,-1,0,1,2,3].collect {|i| 2**i}, gammas = [-2,-1,0,1,2,3].collect {|i| 2**i}, params = {}) ⇒ Object

Perform cross validation search on given gamma/cost values, using an RBF kernel, returning the best performing model and optionally displaying a contour map of performance.

training_set

instance of Problem, used for training

cross_valn_set

instance of Problem, used for evaluating models

costs

array of cost values to search across

gammas

array of gamma values to search across

params

Optional parameters include:

  • :evaluator => Evaluator::OverallAccuracy, the name of the class to use for computing performance

  • :show_plot => false, whether to display contour plot

Returns an instance of Model, the best performing model.



39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
 
# File 'lib/svm_toolkit/svm.rb', line 39

def Svm.cross_validation_search(training_set, cross_valn_set, 
                                 costs = [-2,-1,0,1,2,3].collect {|i| 2**i}, 
                                 gammas = [-2,-1,0,1,2,3].collect {|i| 2**i}, 
                                 params = {})
  evaluator = params.fetch :evaluator, Evaluator::OverallAccuracy
  show_plot = params.fetch :show_plot, false

  fjp = ForkJoinPool.new
  task = CrossValidationSearch.new gammas, costs, training_set, cross_valn_set, evaluator
  results, best_model = fjp.invoke task

  if show_plot
    ContourDisplay.new(costs.collect {|n| Math.log2(n)}, 
                       gammas.collect {|n| Math.log2(n)}, 
                       results)
  end

  return best_model
end